Global Combustion Mechanisms for Use in CFD Modeling underOxy-Fuel Conditions

Jimmy Andersen, Christian Lund Rasmussen, Trine Giselsson, and Peter Glarborg*

Department of Chemical and Biochemical Engineering Technical UniVersity of Denmark, DK-2800 Kgs.Lyngby, Denmark

ReceiVed May 16, 2008. ReVised Manuscript ReceiVed December 16, 2008

Two global multistep schemes, the two-step mechanism of Westbrook and Dryer (WD) and the four-stepmechanism of Jones and Lindstedt (JL), have been refined for oxy-fuel conditions. Reference calculationswere conducted with a detailed chemical kinetic mechanism, validated for oxy-fuel combustion conditions. Inthe modification approach, the initiating reactions involving hydrocarbon and oxygen were retained, whilemodifying the H2-CO-CO2 reactions in order to improve prediction of major species concentrations. Themain attention has been to capture the trend and level of CO predicted by the detailed mechanism as well asthe correct equilibrium concentration. A CFD analysis of a propane oxy-fuel flame has been performed usingboth the original and modified mechanisms. Compared to the original schemes, the modified WD mechanismimproved the prediction of the temperature field and of CO in the post flame zone, while the modified JLmechanism provided a slightly better prediction of CO in the flame zone.

Introduction

Oxy-fuel combustion is a promising technique for separatinggaseous CO2 with the intention of storing it, for instance ingeological reservoirs. The separation of CO2 is facilitated byremoving the atmospheric nitrogen from air before combustion.Recirculated combustion products are used for diluting a pureO2 stream. Due to the higher heat capacity and radiativeproperties of CO2 compared to N2, an increased oxygenconcentration is required to obtain the same thermal conditionsas combustion in air. For natural gas, a mixture of 28% O2 and72% CO2 will result in a temperature field similar to combustionin air.1 Oxy-fuel combustion will eventually result in a flue gasconsisting mainly of CO2 and steam. The flue gas can thenundergo a condensation process to remove the H2O, before apart of this flue gas consisting almost entirely of CO2 isrecirculated, while the remaining part is ready for compressionand storage.2

Computational fluid dynamics (CFD) is becoming an impor-tant industrial tool for trouble-shooting and optimization.However, CFD modeling of industrial combustion applicationsis a computationally demanding task. For this reason, it is oftennecessary to apply simplified reaction mechanisms to reducethe computing effort. Chemistry is often represented by a mixed-is-burned assumption or an assumption of chemical equilibrium.Breussin et al.3 performed a CFD analysis of a pure naturalgas/oxygen flame, and found good predictions for the fluiddynamics, temperature and main chemical species concentrationfields (O2, CO2) using both an Eddy Dissipation/mixed-is-burnedapproach and an EDC/chemical equilibrium approach. Themixed-is-burned model did, however, fail in predicting COproperly. In all the flames predicted, the equilibrium model

showed to be superior to the mixed-is-burned model. Thetemperature and species concentrations predictions were better,since the effect of molecular dissociation has been accountedfor. An alternative to these approaches is to employ finite-ratechemistry, using a scheme consisting of one or several globalreactions. A number of simplified methane oxidation mecha-nisms has been proposed in literature.4-9 Brink et al.10 testedthis approach under perfectly stirred reactor conditions. Anequilibrium approach was compared with a three-step irrevers-ible mechanism,10 the four-step mechanism suggested by Jonesand Lindstedt6 and a detailed mechanism suggested by Glarborget al.11 Their conclusions were that thermodynamic equilibriumprovides a poor description under conditions with a strongcoupling between turbulent mixing and chemical reactions. Theglobal mechanisms resembled the detailed model well at fuel-lean conditions. At fuel-rich conditions, the accuracy of thethree-step irreversible model was less satisfactory, whereas thefour-step mechanism performed better and correctly approachedthe equilibrium composition at long residence times. The four-step mechanism was reported to be a good compromise betweenaccuracy and computational effort.10

Examples of CFD modeling using the global mechanisms canbe found in the literature.10,12-14 Also, in industrial CFDmodeling, the global mechanisms are used frequently, presum-

* To whom correspondence should be addressed. Tel: +45 4525 2840.Fax: +45 4588 2258. E-mail: pgl@kt.dtu.dk.

(1) Andersson, K.; Johnsson, F. Fuel 2007, 86, 656–688.(2) Wall, T. F. Proc. Combust. Inst. 2007, 31, 31–47.(3) Breussin, F.; Lallemant, N.; Weber, R. Combust. Sci. Technol. 2000,

160, 369–397.

(4) Dryer, F. L.; Glassman, I. Proc. Combust. Inst. 1972, 14, 987–1003.(5) Westbrook, C. K.; Dryer, F. L. Combust. Sci. Technol. 1981, 27,

31–44.(6) Jones, W. P.; Lindstedt, R. P. Combust. Flame 1988, 73, 233–249.(7) Dupont, V.; Pourkashanian, M.; Williams, A.; J. Inst. Energy 1993,

66, 20–28.(8) Nicol, D. G.; Malte, P. C.; Hamer, A. J.; Roby, R. J.; Steele, R. C.

J. Eng. Gas Turbines Power 1999, 121, 272–280.(9) Meredith, K. V.; Black, D. L. 44th AIAA Aerospace Sciences Meeting

2006, 19, 14161–14167.(10) Brink, A.; Kilpinen, P.; Hupa, M.; Kjaldman, L. Combust. Sci.

Technol. 1999, 141, 59–81.(11) Glarborg, P.; Alzueta, M. U.; Dam-Johansen, K.; Miller, J. A.

Combust. Flame 1998, 115, 1–27.(12) Brink, A.; Mueller, C.; Kilpinen, P.; Hupa, M. Combust. Flame

2000, 123, 275–279.

Energy & Fuels 2009, 23, 1379–1389 1379

10.1021/ef8003619 CCC: $40.75 2009 American Chemical SocietyPublished on Web 01/28/2009

ably because the models are simple, cheap, and readily available.Furthermore, computationally simple mechanisms do provideadequate results if only the main species concentrations andthe temperature picture is of interest. The simplified schemescannot be expected, however, to work as well under oxy-fuelcombustion conditions, as they do for conventional combustion.The exchange of the largely inert N2 with a chemically reactivecompound, CO2; at least at high temperatures, has been shownto change the importance of some of the elementary reactionsgoverning the combustion,15 thereby requiring a modificationof the global multistep reaction mechanisms to make them validunder oxy-fuel conditions.

The objective of the present work is to modify two simplemultistep mechanisms, used frequently for describing conven-tional combustion, to handle the increased CO2 concentrationunder oxy-fuel conditions. The two-step hydrocarbon oxidationmechanism by Westbrook and Dryer (WD)5 is selected, sincethis scheme is directly available as default in the commercialCFD code Fluent.16 The second scheme selected is the four-step mechanism of Jones and Lindstedt (JL).6 The JL schemeis more complex but also has a higher accuracy and similar tothe WD scheme, it is used regularly in CFD modeling ofindustrial applications.

The selected simplified mechanisms are compared withreference calculations, conducted with the detailed combustionmechanism proposed by Glarborg and Bentzen.15 Their kineticmodel provides good agreement with oxy-fuel combustionexperiments. The work presented herein bears some resemblanceto the work of Brink et al.10 They also tested the Jones andLindstedt four-step mechanism with reference calculations witha detailed mechanism, but only for conventional combustionconditions.

Modeling Approach. The Eddy Dissipation Concept (EDC)is a popular turbulence chemistry interaction model for CFDanalysis of combustion applications. The Eddy DissipationConcept is an extension of the Eddy Dissipation model17 basedon the work by Gran and Magnussen.18,19 In the EDC model,chemical reactions are assumed to occur in the fine structuresof the computational cells. These small scale structures can bepictured as a part of the cell, where Kolmogorov-sized eddiescontaining combustion species are situated so close together thatmixing on the molecular level is taking place. The EDC modelevaluates the volume of each cell, where mixing on a molecularscale is occurring, and treats this part of the cell as a PerfectlyStirred Reactor (PSR/CSTR). The cell volume fraction andreactor residence time is depending on turbulence parametersfor the specific computational cell.18,19

Since the turbulence/chemistry interaction description in CFDmay involve ideal reactor modeling, a comparison of the perfor-mance of different global multistep reaction mechanisms shouldbe conducted under similar conditions. True to the concept of Granand Magnussen,18,19 Brink et al.10 tested the simplified models underperfectly stirred reactor conditions. However, it is questionablewhether commercial CFD codes actually employ a PSR solver forthe chemistry, since the resulting algebraic equations may yieldconvergence problems. More likely, codes like Fluent employ anumerical solver that performs an integration in time. Consequently,in the present work, isothermal plug flow reactor (PFR) modeling,rather than PSR calculations, is used to compare the global and

detailed mechanisms. Detailed model predictions have beenobtained from plug flow simulations using the Senkin code fromthe Chemkin-II library.20 The global multistep model PFRcalculations have been performed in Matlab. Even though theChemkin 4.0 package21 can handle the noninteger reaction ordersthat are often applied in global mechanisms, the Matlab code waspreferred because it facilitated comparison of several parametersfor both the detailed and global models. It should be noted,however, that convergence can be problematic when using non-integer reaction coefficients and because of this, the stiff numericalsolver ode15s was preferred over ordinary ODE solvers.

Detailed Chemical Kinetic Model (DCKM). In this work,the global mechanisms are compared to the detailed chemicalkinetic mechanism (DCKM) presented by Glarborg and Bentz-en.15 To the authors’ knowledge, this mechanism is the onlyone that has been validated against oxy-fuel experiments. Themechanism satisfactorily predicts CO, CO2 and O2 concentra-tions from oxidation of CH4 by an O2/CO2 mixture in a flowreactor with residence times of approximately 1 s.15

The Westbrook and Dryer Two-Step Mechanism (WD).The WD model consists of two reactions, where the last step,oxidation of CO to CO2, is reversible. The mechanism is listedin the form of three irreversible steps,

CH4 + 1.5O2fCO+ 2H2O (WD1)

CO+ 0.5O2fCO2 (WD2)

CO2fCO+ 0.5O2 (WD3)

Table 1 displays the reaction rate data for the WD mechanism.The rate constants for (WD1) and (WD2) originate from Dryerand Glassman4 who studied high temperature oxidation reactionsof carbon monoxide and methane under fuel lean conditions (λ> 2) in a turbulent flow reactor. Later, Westbrook and Dryer5

included the reverse reaction step for CO2 decomposition (WD3)in order to reproduce the proper heat of reaction and pressuredependence of the [CO]/[CO2] equilibrium.

The Jones and Lindstedt Four-Step Mechanism (JL).Jones and Lindstedt6 developed four-step global mechanismsfor several hydrocarbon fuels. For methane, it involves thefollowing steps:

CH4 + 0.5O2fCO+ 2H2 (JL1)

CH4 +H2OfCO+ 3H2 (JL2)

H2 + 0.5O2aH2O (JL3)

CO+H2OaCO2 +H2 (JL4)

The mechanism consists of two irreversible reactions, (JL1)and (JL2), describing the initial oxidation steps of a hydrocarbon.The two reversible reactions, (JL3) and (JL4), control the rateof reaction for CO and H2. The rate coefficients for methanecombustion are displayed in Table 2. Jones and Lindstedtvalidated the model against data for premixed methane andpropane flames along with diffusion flame data for a methane-air flame. The mechanism was reported to perform well for bothfuel-lean and moderately fuel-rich stoichiometries.6 Two setsof rate parameters for reaction (JL3) were presented. Set (JL3a)was the preferred expression, but since it involved negative

Table 1. Westbrook and Dryer Global Multi Step Methane Combustion Mechanism with Kinetic Data (units in cm, s, cal, and mol)

mechanism reactions A � Ea reaction orders ref

WD1 CH4 + 1.5 O2 f CO + 2 H2O 1.59 × 1013 0 47.8 × 103 [CH4]0.7[O2]0.8 4WD2 CO + 0.5 O2 f CO2 3.98 × 1014 0 40.7 × 103 [CO][O2]0.25[H2O]0.5 4WD2r CO2 f CO + 0.5 O2 5.0 × 108 0 40.7 × 103 [CO2] 5

1380 Energy & Fuels, Vol. 23, 2009 Andersen et al.

reaction orders that might cause numerical problems, the setfor (JL3b) was proposed as an alternative.6

If (JL3) had been an elementary reaction, the reverse rateconstant could be determined from expression 1,

RJL3,f

RJL3,r

)KJL3 )kJL3,f[H2][O2]

0.5

kJL3,r[H2O](1)

Here, KJL3 is the equilibrium constant, kJL3 is the rate constant,and the f and r subscribts refer to the forward and reversereactions, respectively. However, (JL3) is a global reaction andthe forward reaction orders do not follow the stoichiometry.For this reason, the reverse reaction order for reaction (JL3a)or (JL3b) cannot be derived according to eq 1. In order for anequilibrium approach to be maintained for a global reaction,the expression 1 must still hold at equilibrium, here exemplifiedwith reaction (JL3b):

RJL3,f

RJL3,r

)RJL3b,f

*

RJL3b,r*

)kJL3,f[H2][O2]

0.5

kJL3,r[H2O])

kJL3b,f* [H2]

0.25[O2]1.5

kJL3b,r* [H2O]

(2)

The superscript * refers to the global rate expressions. Sincethe forward rate constant for the global and the stoichiometricexpression, respectively, must be identical (kJL3, f ) kJL3b, f

* ), theconcentration dependence of the reverse reactions can be foundas follows:

kJL3b,r* ) kJL3,r[H2]

-0.75[O2]w RJL3b,r* ) kJL3,r[H2]

-0.75[O2][H2O]

(3)

In the present work, the derivation of the backward rates is doneby evaluating the forward rates divided by the equilibriumconstant at a series of temperatures (from 500 to 2500 K with100 K increments) and then fitting an Arrhenius expression tothe results, to obtain an individual expression for the reversereaction. For reaction (JL3a) we obtain the expression,

kJL3a,r* (T)) 2 . 6 × 1018 · T-0.877 · exp(-49260

T ) (4)

and for reaction (JL3b),

kJL3b,r* (T)) 7.1 × 1017 · T-0.877 · exp(-49260

T ) (5)

These expressions make it possible to implement the forwardand reverse reactions as irreversible steps, facilitating the usein commercial CFD software such as Fluent.16

In the following, only reaction 3b of the JL mechanism isused. Attempts to use 3a resulted in problems with convergencewhich may limit its use in CFD. Another concern related to 3ais that the reverse rate for (JL3a) is independent of the H2Oconcentration. This makes sense when the purpose of thereaction is to dampen the forward reaction, but under conditionswith large H2O concentrations or even H2O in the oxidizerstream a zero reaction order may be inappropriate.

A way to implement the JL mechanism in Fluent16 is toimport it as a Chemkin input file. In the newer versions ofChemkin, the FORD keyword is used to overwrite the forwardreaction order, allowing global expressions to be implemented.Table 3 shows an example of a Chemkin input file, with reaction(3b) implemented. Note that in the reverse H2-O2 reaction, H2

and O2 have been added as reactants with 0 as stoichiometrycoefficients. This is required to allow the program to overwritethe forward reaction orders.

Results and Discussion

In this work, we evaluate the performance of the WD and JLschemes under conditions of conventional combustion and oxy-fuel combustion, respectively, assuming plug-flow. Then the twoschemes are revised for use under oxy-fuel conditions and testedagain against reference calculations with the detailed mechanism.Finally, the original and modified schemes are implemented andcompared for CFD calculations of a turbulent diffusion propane/O2/CO2 flame under conditions similar to those reported recentlyby Andersson and Johnsson.1

Performance of the WD and JL Mechanisms. Both the JLand the WD global mechanisms have been used extensively inCFD models for conventional combustion in air. Before weinvestigate how the schemes function under oxy-fuel conditions,we test them under normal combustion conditions by comparingwith reference calculations with the detailed reaction mechanism.Since the mechanisms were optimized for fuel-lean conditions,we evaluate them at conditions with excess air.

Figure 1 compares CO, O2 and CO2 concentrations in anisothermal plug flow reactor at 1600 K and an excess air ratio

Table 2. Jones Lindstedt Global Multi-Step Methane Combustion Mechanism with the Kinetic Rate Data (units in cm, s, cal, and mol)

no. reactions A � Ea reaction orders ref

JL1 CH4 + 0.5 O2 f CO + 2 H2 7.82 × 1013 0 30.0 × 103 [CH4]0.5[O2]1.25 6JL2 CH4 + H2O f CO + 3 H2 0.30 × 1012 0 30.0 × 103 [CH4][H2O] 6JL3a H2 + 0.5 O2 h H2O 4.45 × 1018 -1 40.0 × 103 [H2]0.5[O2]2.25[H2O]-1 6JL3b H2 + 0.5 O2 h H2O 1.21 × 1018 -1 40.0 × 103 [H2]0.25[O2]1.5 6JL4 CO + H2O h CO2 + H2 2.75 × 1012 0 20.0 × 103 [CO][H2O] 6

Table 3. Chemkin-Code for the JL Mechanism (3b H2-O2 Reaction) (units: cm, s, cal, mol, and K)

ELEMENTS C O H N END

SPECIES CH4 O2 H2O N2 CO CO2 H2 ENDREACTIONSCH4 + 0.5 O2 ) > CO+ 2H2 7.82 × 1013 0 30.0 × 103 !Jones Lindstedt 88FORD /CH4 0.5/FORD /O2 1.25/CH4 + H2O ) > CO + 3H2 0.30 × 1012 0 30.0 × 103 !Jones Lindstedt 88H2 + 0.5 O2 ) > H2O 1.209 × 1018 -1 40.0 × 103 !Jones Lindstedt 88FORD /H2 0.25/FORD /O2 1.5/H2O + 0 O2 +0 H2 ) > H2 + 0.5 O2 7.06 × 1017 -0.877 97.9 × 103 !CalculatedFORD /H2 -0.75/FORD /O2 1/FORD /H2O 1/CO + H2O ) CO2+H2 0.275 × 1013 0 20.0 × 103 !Jones Lindstedt 88END

Global Combustion Mechanisms Energy & Fuels, Vol. 23, 2009 1381

of λ ) 1.2. These results, along with other simulations, showthat both the WD and JL mechanisms adequately predict O2

and CO2 levels at fuel lean conditions. Thereby, they wouldalso yield a satisfactory prediction of the heat release fornonisothermal conditions at these stoichiometries. Only the JLmechanism predicts CO correctly at longer times under fuellean conditions. The WD mechanism tends to overpredict theCO exit concentration and may not be sufficiently accurate forCO emission modeling.

As expected, the results show differences in the predictedignition time between the three models. Only the detailed modelcan describe the slow build-up of the radical pool that eventuallylead to ignition in a plug-flow calculation. The global schemescannot account for an ignition delay. The WD mechanism wasdeveloped to describe the postignition fuel-consumption ratein PFR experiments, whereas the JL mechanism was optimizedfor flame conditions, where a radical pool is available throughdiffusional processes. It is difficult to assess how inaccuraciesin the description of ignition affect a CFD calculation. In theEDC approach, PSR/PFR reactor residence times may be in therange of 10-4 to 10-3 s, i.e., time scales where there areconsiderable differences between modeling predictions. How-ever, all three models predict the time scale for complete

conversion of CH4 to CO2 to be of the order of 10-3 s, whichis satisfactory.

The performances of the WD and JL mechanisms under fuel-lean oxy-firing conditions are displayed in Figures 2 and 3. Atlow temperatures (<1300 K) consumption of CO2 is practicallynegligible,15 and CO2 is expected to have little chemical effectunder these conditions. At higher temperatures, such as in thefigures, atomic hydrogen may start to convert CO2 to CO,resulting in a change in both the CO/CO2 ratio and thecomposition of the O/H radical pool compared to conventionalcombustion. In general, the level of agreement between theglobal mechanisms and the reference calculations is similar tothat observed under conventional combustion conditions. Asexpected, the WD mechanism cannot predict the CO exitconcentration accurately; differences are seen at all temperatures.With the exception of the ignition timing, predictions of the JLmechanism generally compare well with those of the detailedmechanism. The ability of the JL scheme to compensate forthe water-gas shift reaction (JL4) allows it to capture mostchanges caused by the elevated CO2 concentration. Conse-quently, it predicts correct levels for all major species at longerreaction times, even under fuel-rich conditions. However, theCO peak levels are overpredicted in most cases.

Figure 1. Major species concentrations in plug flow calculations. Comparison between the DCKM, the WD and the JL mechanisms at λ ) 1.2 and1600 K under “air” firing conditions (21% O2 and 79% N2).

Figure 2. Major species concentrations in plug flow calculations. Comparison between the DCKM, the WD, and the JL mechanisms at λ ) 1.2 and1200 K (top), 1600 K (middle), and 2000 K (bottom) under “oxy”-firing conditions (28% O2 and 72% CO2).

1382 Energy & Fuels, Vol. 23, 2009 Andersen et al.

Refined Schemes for Oxy-Fuel Combustion. It is apparentthat the WD mechanism needs improvement to become ap-plicable for oxy-fuel conditions due to the poor CO prediction.The JL mechanism works satisfactorily under oxy-fuel condi-tions, even though the prediction of the peak CO levels couldbe improved. In the present work, both global models aremodified to improve the prediction of the steady state concentra-tions and the CO trends compared to the reference calculations.

The global mechanisms, refined for oxy-fuel conditions, aresummarized in Table 4. The following criteria were employedin the modification procedure:

• In time, the concentrations should approach correctly thechemical equilibrium values.

• The peak CO concentrations should approximate the valuespredicted by the detailed model.

The modification window involved temperatures of 1200-2000K and stoichiometries in the range 0.8 < λ < 1.5.

In the WD scheme, the major issue was the rate coefficientsfor reaction (WD3), which did not secure an approach toequilibrium values for CO and CO2. These were consequentlymodified by applying the global mechanism equilibrium ap-proach for the CO-CO2 reaction, following the procedurediscussed above (see eqs 2-5). This change caused the WDscheme to predict the approach to equilibrium correctly, butresulted in an underprediction of CO at higher temperatures.To improve the CO prediction, it was necessary also to modify(WD2). Consequently, both the A-factor and the activation

energy for (WD2) was lowered to match better the detailedmodel predictions.

When comparing the CO levels predicted by the JL mech-anism with the detailed mechanism, it was observed that besidesthe difference in ignition timing, the peak CO levels deviated,most pronounced under oxy-fuel conditions (Figures 2 and 3).The predicted CO oxidation rate in the JL scheme is governedby the reaction with water vapor (JL4). Examination of themechanism showed that it was the availability of H2O (formedin reaction JL3), as well as the rate constant for (JL4), thatlimited the CO oxidation rate. Attempts to modify the JL4 rateconstant proved unsuccesful; a decrease in kJL4 resulted inundesired changes at some conditions (prolonged burnoutperiod), whereas an increase had little effect, since the formation

(13) Mueller, C.; Brink, A.; Kilpinen, P.; Hupa, M.; Kremer, H. CleanAir 2002, 3, 145–163.

(14) Saario, A.; Oksanen, A. Energy Fuels 2008, 22, 297–305.(15) Glarborg, P.; Bentzen, L. L. B. Energy Fuels 2007, 22, 291–296.(16) Fluent 6.2 users guide, Fluent Inc., Centerra Resource Park, 10

Cavendish Court, Lebanon, NH 03766, 2005.(17) Magnussen, B. F.; Hjertager, B. H. Proc. Combust. Inst. 1971, 13,

649–657.(18) Gran, I.; Magnussen, B. F. Combust. Sci. Technol. 1996, 119, 171–

190.(19) Gran, I.; Magnussen, B. F. Combust. Sci. Technol. 1996, 119, 191–

217.(20) Lutz, A. E.; Kee, R. J.; Miller, J. A.;Senkin: A Fortran Program

for Predicting Homogeneous Gas Phase Chemical Kinetics With SensitiVityAnalysis, Report No. SAND87-8248 Laboratories, 1990.

Figure 3. Major species concentrations in plug flow calculations. Comparison between the DCKM, the WD and the JL mechanisms at 1600 K andλ ) 0.8 (top), λ ) 1.0 (middle), and λ ) 1.2 (bottom) under “oxy”-firing conditions (28% O2 and 72% CO2).

Table 4. Modified Multi Step Methane Combustion Mechanisms with Kinetic Rate Data - units in cm, s, cal, mol

reaction number reactions A � Ea reaction orders

WD1 CH4 + 1.5 O2 f CO + 2 H2O 1.59 × 1013 0 47.8 × 103 [CH4]0.7[O2]0.8

WD2(modified) CO + 0.5 O2 f CO2 3.98 × 108 0 10.0 × 103 [CO][O2]0.25[H2O]0.5

WD3(modified) CO2 f CO + 0.5 O2 6.16 × 1013 -0.97 78.4 × 103 [CO2][H2O]0.5[O2]-0.25

JL1 CH4 + 0.5 O2 f CO + 2 H2 7.82 × 1013 0 30.0 × 103 [CH4]0.5[O2]1.25

JL2 CH4 + H2O f CO + 3 H2 3.00 × 1011 0 30.0 × 103 [CH4][H2O]JL3(modified) H2 + 0.5 O2 f H2O 5.0 × 1020 -1 30.0 × 103 [H2]0.25[O2]1.5

JL3 reverse H2O f H2 + 0.5 O2 2.93 × 1020 -0.877 97.9 × 103 [H2]-0.75[O2][H2O]JL4 CO + H2O a CO2 + H2 2.75 × 1012 0 20.0 × 103 [CO][H2O]

Global Combustion Mechanisms Energy & Fuels, Vol. 23, 2009 1383

of H2O (through JL3) became rate limiting. Consequently,predictions with the JL scheme were improved by modifyingthe rate coefficients for JL3, increasing the pre-exponential factorand decreasing the activation energy.

The parameters for the fuel-specific reactions were not partof the modification procedure; the kinetic data for theinitiating reactions, i.e., (JL1), (JL2) and (WD1), were allretained. This is consistent with the flow reactor results ofGlarborg and Bentzen15 that indicate that the temperature forthe initiation of reaction are very similar for reactive flowswith and without CO2. This implies that the main chemicaldifference between combustion in O2/N2 and O2/CO2 relatesto the CO/H2 reaction subset. Since the fuel-specific stepshave been retained, the proposed schemes can easily be

modified to describe higher hydrocarbons by adopting theappropriate rate data from the original mechanisms.4,6

However, the inability of the global schemes to predictignition timing has not been addressed.

PFR Tests of the Refined Schemes. The impact of themodifications on the CO prediction is illustrated in Figure 4for lean conditions and 1600 K. The modification of the WDmechanism results in an improved prediction of the exit COconcentration, while the peak value is slightly lowered. Themodified JL mechanism predicts a reduced peak CO level,due to the increased value of kJL3. The major speciesconcentrations predicted by the modified mechanisms aredisplayed in Figures 5 and 6. The results confirm that themodified schemes constitute an improvement over the WDand JL mechanisms for oxy-fuel conditions. Compared tothe original schemes, the prediction in particular of the COconcentration has been improved. The improvement is mostpronounced for the WD mechanism, which now predicts boththe trend in CO and the approach to equilibrium concentra-tions reasonably well. For the modified JL mechanism, thepredictions of the CO trend and the peak CO levels havebeen improved. Under fuel-rich conditions, both modifiedmechanisms approach steady-state concentrations reasonablywell, as shown in Figure 6. This indicates that the schemesmay be applied also for modeling staged or slightly under-stoichiometric combustion applications.

Figure 7 shows the performance of the modified globalmechanisms when simulating a recirculated flue gas containingH2O as well as CO2; an option considered for oxy-fuel-firedpower plants. Calculations were conducted for an oxidizerstream consisting of 28% O2, 40% CO2 and 32% H2O insteadof the 28% O2 and 72% CO2. The results confirm that themodified schemes predict the major species concentrations fairlywell also under these conditions.

Figure 4. CO concentrations in plug flow calculations. Comparisonbetween the DCKM, WD, JL, and modified mechanisms at 1600 Kand λ ) 1.2 under “oxy”-firing conditions (28% O2 and 72% CO2).

Figure 5. Major species concentrations in plug flow calculations. Comparison between the DCKM, the modified WD and the modified JL mechanismsat λ ) 1.2 and 1200 K (top), 1600 K (middle), and 2000 K (bottom) under “oxy”-firing conditions (28% O2 and 72% CO2).

1384 Energy & Fuels, Vol. 23, 2009 Andersen et al.

Figure 6. Major species concentrations in plug flow calculations. Comparison between the DCKM, the modified WD and the modified JL mechanismsat 1600 K and λ ) 0.8 (top), λ ) 1.0 (middle), and λ ) 1.2 (bottom) under “oxy”-firing conditions (28% O2 and 72% CO2).

Figure 7. Major species concentrations in plug flow calculations at 1600 K. Comparison between the modified WD, JL, and detailed mechanismsunder “oxy”-firing conditions with an oxidizer stream of 28% O2, 40% CO2 and 32% H2O.

Global Combustion Mechanisms Energy & Fuels, Vol. 23, 2009 1385

The turbulent mixing time scales applied in the individualCFD cells are of the order of 0.1-100 ms.22 As expected,the modified mechanisms do not match the detailed mech-anism on the smaller timescales. This implies that the globalschemes have a limited accuracy in describing changesoccurring on a small time scale or on an individual cell basisin a CFD computation. However, in terms of larger time

scales or larger cell areas, the global mechanisms are capableof predicting satisfactorily the heat release and major speciesconcentrations, provided the turbulence modeling providesaccurate turbulence level predictions as input for the EDCturbulence chemistry interaction model.

CFD Modeling with the Refined Schemes. Recently,Andersson and Johnsson1 reported measurements in a turbulentdiffusion flame operated under oxy-fuel conditions. The ex-perimental setup consisted of a 100 kW down-fired refractorylined furnace, with a swirl burner configuration. The fuel usedwas propane and the oxy-fuel experiments (27% oxygen in CO2)were compared with data obtained in air. As part of theevaluation of the global schemes, we have conducted a CFDanalysis the oxy-fuel flame. The geometry of the setup wassimplified to a 2D case, even though the setup was not entirelyaxi-symmetric, due to four cooling tubes near the furnace walls.The heat loss to the cooling tubes was accounted for by applyinga piecewise constant wall temperature to obtain gas temperaturesnear the walls in agreement with measurements. The grid wasconstructed with ∼30.000 cells. A grid-independency checkindicated that this grid size was sufficient.

The commercial CFD code Fluent was used for the calcula-tion. The realizable k-ε turbulence model was adopted alongwith the P1 radiation model. Second-order upwind discretizationwas used for all transported scalars. The CFD settings aresummarized in Table 5. Both the Eddy Dissipation Model(EDM) and the Eddy Dissipation Concept (EDC) were appliedin the modeling of the oxy-fuel flame. We recommend to usethe EDC turbulence chemistry interaction model, when modelingoxy-fuel flames. The traditional “mixed-is-burned” approachoffered by the EDM is not likely to be applicable in reactionsystems where reverse reactions play an important role, as isthe case under oxy-fuel conditions due to CO2 decompositionat high temperatures. Limitations of the EDM model arediscussed in more detail by Brink et al.12

The CFD results presented here were all performed usingthe EDC turbulence-chemistry interaction model, applying boththe original WD and JL mechanisms and the modified versionssuggested in Table 4. Since the fuel used in the experiments ispropane, rate data and stoichiometric relationships for theinitiating reactions (JL1, JL2, and WD1, see Table 4) weremodified according to the original references.4,6 These reactionsare identical in the original and modified versions.

(21) Kee, R. J.; Rupley, F. M.; Miller, J. A.; Coltrin, M. E.; Grcar, J. F.;Meeks, E.; Moffat, H. K.; Lutz, A. E.; Dixon-Lewis, G.; Smooke, M. D.;Warnatz, J.; Evans, G. H.; Larson, R. S.; Mitchell, R. E.; Petzold, L. R.;Reynolds, W. C.; Caracotsios, M.; Stewart, W. E.; Glarborg, P.; Wang, C.;Adigun, O.; Houf, W. G.; Chou, C. P.; Miller, S. F.; Ho, P.; Young, D. J.;CHEMKIN Release 4.0, Reaction Design, Inc., San Diego, CA ( 2004).

(22) Kjaldman, L.; Brink, A.; Hupa, M. Combust. Sci. Technol. 2000,154, 207–227.

Figure 8. Temperature [K] Left: at 21.5 cm downstream of the burner. Right: at 38.4 cm downstream of the burner. Dots: Experimental data fromAndersson and Johnsson.1 Dashed line: Modified WD mechanism. Solid line: Original WD mechanism.

Figure 9. Temperature [K] Axial temperature at the centerline of thesetup. Dots: Experimental data from Andersson and Johnsson.1 Dashedline: Modified WD mechanism. Solid line: Original WD mechanism.

Table 5. Submodels and settings for the CFD analysis

grid turbulence radiationchemistryinteraction mechanisms

2D30.000 cells

realizablek-ε

P1 Eddy Dissipationconcept

modified(Table 3)original JL andWD (Tables 1,2)

1386 Energy & Fuels, Vol. 23, 2009 Andersen et al.

A satisfactory agreement between major species concentra-tions predicted by the CFD model and experimental results

was achieved for the air case. Here, only results from theoxy-fuel case is presented. It should be noted that deviations

Figure 10. Radial CO concentration (%dry) Left: at 21.5 cm downstream of the burner. Right: at 38.4 cm downstream of the burner. Dots: Experimentaldata from Andersson and Johnsson.1 Dashed line: Modified WD mechanism. Solid line: Original WD mechanism.

Figure 11. Radial O2 concentration (%dry) Left: at 21.5 cm downstream of the burner. Right: at 38.4 cm downstream of the burner. Dots: Experimentaldata from Andersson and Johnsson.1 Dashed line: Modified WD mechanism. Solid line: Original WD mechanism.

Figure 12. Temperature [K] Left: at 21.5 cm downstream of the burner. Right: at 38.4 cm downstream of the burner. Dots: Experimental data fromAndersson and Johnsson.1 Dashed line: Modified JL mechanism. Solid line: Original JL mechanism.

Global Combustion Mechanisms Energy & Fuels, Vol. 23, 2009 1387

between the CFD predictions and the experimental resultscan only partly be attributed to an insufficient combustionmechanism. A satisfactory prediction of turbulence and flow

field is essential; unfortunately, no data for velocity andvelocity fluctuations are available from the oxy-fuel experi-ments. Radiative properties are also affected when changingfrom air to oxy-fuel combustion, and it is likely thatdifferences in the temperature profiles are due to insufficientradiative modeling along with the simplifications in geometry(neglection of the cooling tubes). Furthermore, soot forma-tion, which is important for the flame radiation, was notaccounted for in the modeling.

Figures 8 and 9 show comparison between experimentalresults1 and modeling predictions with the WD schemes forthe radial and centerline temperature profiles, respectively,while Figures 10 and 11 show results for radial CO and O2

concentrations. The CO and O2 concentrations are shown fortwo measurement planes, 21.5 and 38.4 cm downstream ofthe burner. Solid lines denote calculations with the originalWD scheme, while dashed lines donte predictions with themodified WD scheme. In general, the modified WD mech-anism provides an improved prediction of both the temper-ature and the postflame conditions and emissions, comparedto the original scheme. The modified mechanism predicts wellthe CO levels in the region outside the flame zone (Figure10). The improvement can be attributed to the new equilib-

Figure 13. Temperature [K] Axial temperature at the centerline of thesetup. Dots: Experimental data from Andersson and Johnsson.1 Dashedline: Modified JL mechanism. Solid line: Original JL mechanism.

Figure 14. Radial CO concentration (%dry) Left: at 21.5 cm downstream of the burner. Right: at 38.4 cm downstream of the burner. Dots: Experimentaldata from Andersson and Johnsson.1 Dashed line: Modified JL mechanism. Solid line: Original JL mechanism.

Figure 15. Radial O2 concentration (%dry) Left: at 21.5 cm downstream of the burner. Right: at 38.4 cm downstream of the burner. Dots: Experimentaldata from Andersson and Johnsson.1 Dashed line: Modified JL mechanism. Solid line: Original JL mechanism.

1388 Energy & Fuels, Vol. 23, 2009 Andersen et al.

rium fit for the WD3 reaction. The modified model alsopredicts a different CO level in the flame zone, but due tothe limited experimental data points for CO in this region,both sets of predictions are considered to be within experi-mental uncertainty. The differences in the predicted CO levelsdo, however, emphasize the importance of the chemicalmechanism in the CFD computation. In the very fuel-richparts of the flame, the CO levels are underpredicted in thisCFD analysis. This indicates that it may be necessary to applya more complex model than the WD scheme for reliablemodeling of fuel-rich regions of a combustion system.

Comparisons for temperature profiles (Figures 12 and 13) andCO/O2 concentrations prediction (Figures 14 and 15) have alsobeen conducted for the JL schemes. Figures 14 and 15 compareCO and O2 concentrations from the original JL mechanism andthe modified mechanism with the experimental measurements attwo measurement planes 21.5 and 38.4 cm downstream of theburner. From Figure 14, it can be seen that the modified JLmechanism predicts a lower CO concentration in the center of thefurnace inside the flame, improving the agreement with experiment.The prediction of a lower CO level by the modified scheme isconsistent with the PFR calculations discussed above. The modi-fications in the JL mechanism also induce changes in the temper-ature predictions (Figures 12 and 13) and the O2 prediction (Figure15), but whether these changes constitute improvements over theoriginal mechanism cannot be concluded.

Conclusions

Two global multistep mechanisms, the two-step mechanismby Westbrook and Dryer and the four-step mechanism by Jonesand Lindstedt, have been tested and refined for oxy-fuelconditions, based on comparison with model predictions witha detailed chemical kinetic mechanism. In the modification, the

initiation reactions involving the hydrocarbon fuel were unal-tered. Changes were made in the CO-CO2 reaction subset toobtain an improved fit for CO levels and steady state emissionsas predicted by the detailed mechanism.

The modified schemes provide improved agreement with thedetailed mechanism compared to the original mechanisms forisothermal plug flow reactor modeling under oxy-fuel combus-tion conditions. The improvement was most pronounced for WDmechanism, where the modified scheme yielded a betterprediction of the peak and exit concentrations of CO. Themodified JL mechanism offers a slight improvement in predict-ing CO trends.

A CFD analysis of a propane oxy-fuel flame was performedusing both the original and modified mechanisms. Themodified JL mechanism performed slightly better than theoriginal, regarding CO predictions in the flame zone. Themodified WD mechanism improved the prediction of COespecially in the post flame zone, but also a better temperatureagreement with experimental data was obtained. In general,it is recommended to use the EDC turbulence chemistryinteraction model, when modeling oxy-fuel flames. Thetraditional “mixed is burned” approach offered by the EddyDissipation Model is not likely to be applicable in acombustion system where the CO2 dissociation can have asignificant chemical effect.

Acknowledgment. This work was funded by Vattenfall Researchand Development AB with Karin Eriksson as project manager. Theauthors would like to thank Anders Brink and W.P. Jones for helpfuldiscussions.

EF8003619

Global Combustion Mechanisms Energy & Fuels, Vol. 23, 2009 1389